Rates of large breast cancers should decrease in a population that is offered mammography screening, but women continue to present with them. We sought an explanation in a population-based epidemiological study of 1,459 women diagnosed with invasive breast cancer in 2002-2003 in Australia; breast cancers were > or =2 cm in 766 women (53%) and 11-1.9 cm in a comparison group (693, 47%). We interviewed the women about their personal, mammogram and breast histories in the years before diagnosis and collected biological characteristics of tumors and mammogram dates from medical records. The strongest correlate of breast cancer size at diagnosis was the method of detection: the odds of a > or =2 cm breast cancer was substantially lower for detection by a screening mammogram (OR = 0.27, 95% CI 0.21-0.34; p < 0.001) than for detection after a breast symptom. Higher BMI (ORs approximately 1.6 for > or =25 kg/m(2)), higher cancer grade (ORs of 1.6 for moderate, 2.89 for high grade) and lobular type (OR 2.09, 95% CI 1.45-3.0) were also independent correlates (p < 0.001) of a > or =2 cm breast cancer. HRT use strongly reduced the odds but only in cancers detected after a breast symptom (OR = 0.49, 95% CI 0.33-0.74; p = 0.002), not in those detected by a screening mammogram. As assessed from their proportional contribution to > or =2 cm breast cancers in our study population, lack of mammogram detection, BMI > or =25 kg/m(2) and moderate or high grade of the cancer were the most important factors with population attributable fractions of 42%, 11% and 29% respectively; the first 2 are amenable to intervention. (c) 2008 Wiley-Liss, Inc.